Search

This work is focused on the simulation of a complete hybrid bus vehicle model performing a real-world driving cycle. The simulation framework consists of a coupled co-simulation environment, where all the vehicle sub-system models are linked to achieve a real time exchange of input and output signals. In the vehicle model also the electric devices of the powertrain and accumulation system are included. This co-simulation platform is applied to investigate the hybridization of a 12-m city bus, performing a typical urban driving mission. A comparison between the conventional powertrain is performed against the hybridized version, to highlight the advantages and challenges. In particular, the novelty of this modeling approach is that the IC engine simulation does not rely on pre-processed look-up tables, but exploits a high-fidelity one-dimensional thermo-fluid dynamic model. However, it was necessary to develop a fast simulation methodology to exploit this predictive tool, achieving a low computational cost. The 1D engine model is first validated against the experimental engine map data available, showing a good model predictivity. Then the 1D engine model and the other models of the powertrain are coupled to the vehicle model, in order to follow the prescribed velocity profile of the driving cycle. The complete model is applied under different conditions, to evaluate the impact on performance and emissions and assess the simulation predictivity. [ABSTRACT FROM AUTHOR]

The prediction of internal combustion engine performance and emissions in real driving conditions is getting more and more important due to the upcoming stricter regulations. This work aims at introducing and validating a new transient simulation methodology of an ICE coupled to a hybrid architecture vehicle, getting closer to real-time calculations. A one-dimensional computational fluid dynamic software has been used and suitably coupled to a vehicle dynamics model in a user function framework integrated within a Simulink® environment. A six-cylinder diesel engine has been modeled by means of the 1D tool and cylinder-out emissions have been compared to experimental data. The measurements available have been used also to calibrate the combustion model. The developed 1D engine model has been then used to perform driving cycle simulations considering the vehicle dynamics and the coupling with the energy storage unit in the hybrid mode. The map-based approach along with the vehicle simulation tool has also been used to perform the same simulation and the two results are compared to evaluate the accuracy of each approach. In this framework, to achieve the best simulation performance in terms of computational time over simulated time ratio, the 1D engine model has been used in a configuration with a very coarse mesh. Results have shown that despite the high mesh spacing used the accuracy of the wave dynamics prediction was not affected in a significant way, whereas a remarkable speed-up factor was achieved. This means that a crank angle resolution approach to the vehicle simulation is a viable and accurate strategy to predict the engine emission during any driving cycle with a computation effort compatible with the tight schedule of a design process. [ABSTRACT FROM AUTHOR]

We explore new and existing convolutional neural network (CNN) architectures for sea ice classification using Sentinel-1 (S1) synthetic aperture radar (SAR) data by investigating two key challenges: binary sea ice versus open-water classification, and a multi-class sea ice type classification. The analysis of sea ice in SAR images is challenging because of the thermal noise effects and ambiguities in the radar backscatter for certain conditions that include the reflection of complex information from sea ice surfaces. We use manually annotated SAR images containing various sea ice types to construct a dataset for our Deep Learning (DL) analysis. To avoid contamination between classes we use a combination of near-simultaneous SAR images from S1 and fine resolution cloud-free optical data from Sentinel-2 (S2). For the classification, we use data augmentation to adjust for the imbalance of sea ice type classes in the training data. The SAR images are divided into small patches which are processed one at a time. We demonstrate that the combination of data augmentation and training of a proposed modified Visual Geometric Group 16-layer (VGG-16) network, trained from scratch, significantly improves the classification performance, compared to the original VGG-16 model and an ad hoc CNN model. The experimental results show both qualitatively and quantitatively that our models produce accurate classification results. [ABSTRACT FROM AUTHOR]

Spectral unmixing (SU) aims at decomposing the mixed pixel into basic components, called endmembers with corresponding abundance fractions. Linear mixing model (LMM) and nonlinear mixing models (NLMMs) are two main classes to solve the SU. This paper proposes a new nonlinear unmixing method base on general bilinear model, which is one of the NLMMs. Since retrieving the endmembers' abundances represents an ill-posed inverse problem, prior knowledge of abundances has been investigated by conceiving regularizations techniques (e.g., sparsity, total variation, group sparsity, and low rankness), so to enhance the ability to restrict the solution space and thus to achieve reliable estimates. All the regularizations mentioned above can be interpreted as denoising of abundance maps. In this paper, instead of investing effort in designing more powerful regularizations of abundances, we use plug-and-play prior technique, that is to use directly a state-of-the-art denoiser, which is conceived to exploit the spatial correlation of abundance maps and nonlinear interaction maps. The numerical results in simulated data and real hyperspectral dataset show that the proposed method can improve the estimation of abundances dramatically compared with state-of-the-art nonlinear unmixing methods. [ABSTRACT FROM AUTHOR]

Convolutional neural networks (CNNs) have been widely applied in hyperspectral imagery (HSI) classification. However, their classification performance might be limited by the scarcity of labeled data to be used for training and validation. In this paper, we propose a novel lightweight shuffled group convolutional neural network (abbreviated as SG-CNN) to achieve efficient training with a limited training dataset in HSI classification. SG-CNN consists of SG conv units that employ conventional and atrous convolution in different groups, followed by channel shuffle operation and shortcut connection. In this way, SG-CNNs have less trainable parameters, whilst they can still be accurately and efficiently trained with fewer labeled samples. Transfer learning between different HSI datasets is also applied on the SG-CNN to further improve the classification accuracy. To evaluate the effectiveness of SG-CNNs for HSI classification, experiments have been conducted on three public HSI datasets pretrained on HSIs from different sensors. SG-CNNs with different levels of complexity were tested, and their classification results were compared with fine-tuned ShuffleNet2, ResNeXt, and their original counterparts. The experimental results demonstrate that SG-CNNs can achieve competitive classification performance when the amount of labeled data for training is poor, as well as efficiently providing satisfying classification results. [ABSTRACT FROM AUTHOR]

The trade-off between spatial and temporal resolution limits the acquisition of dense time series of Landsat images, and limits the ability to properly monitor land surface dynamics in time. Spatiotemporal image fusion methods provide a cost-efficient alternative to generate dense time series of Landsat-like images for applications that require both high spatial and temporal resolution images. The Spatial and Temporal Reflectance Unmixing Model (STRUM) is a kind of spatial-unmixing-based spatiotemporal image fusion method. The temporal change image derived by STRUM lacks spectral variability and spatial details. This study proposed an improved STRUM (ISTRUM) architecture to tackle the problem by taking spatial heterogeneity of land surface into consideration and integrating the spectral mixture analysis of Landsat images. Sensor difference and applicability with multiple Landsat and coarse-resolution image pairs (L-C pairs) are also considered in ISTRUM. Experimental results indicate the image derived by ISTRUM contains more spectral variability and spatial details when compared with the one derived by STRUM, and the accuracy of fused Landsat-like image is improved. Endmember variability and sliding-window size are factors that influence the accuracy of ISTRUM. The factors were assessed by setting them to different values. Results indicate ISTRUM is robust to endmember variability and the publicly published endmembers (Global SVD) for Landsat images could be applied. Only sliding-window size has strong influence on the accuracy of ISTRUM. In addition, ISTRUM was compared with the Spatial Temporal Data Fusion Approach (STDFA), the Enhanced Spatial and Temporal Adaptive Reflectance Fusion Model (ESTARFM), the Hybrid Color Mapping (HCM) and the Flexible Spatiotemporal DAta Fusion (FSDAF) methods. ISTRUM is superior to STDFA, slightly superior to HCM in cases when the temporal change is significant, comparable with ESTARFM and a little inferior to FSDAF. However, the computational efficiency of ISTRUM is much higher than ESTARFM and FSDAF. ISTRUM can to synthesize Landsat-like images on a global scale. [ABSTRACT FROM AUTHOR]

In this paper, a novel convolutional neural network (CNN)-based architecture, named fine segmentation network (FSN), is proposed for semantic segmentation of high resolution aerial images and light detection and ranging (LiDAR) data. The proposed architecture follows the encoder–decoder paradigm and the multi-sensor fusion is accomplished in the feature-level using multi-layer perceptron (MLP). The encoder consists of two parts: the main encoder based on the convolutional layers of Vgg-16 network for color-infrared images and a lightweight branch for LiDAR data. In the decoder stage, to adaptively upscale the coarse outputs from encoder, the Sub-Pixel convolution layers replace the transposed convolutional layers or other common up-sampling layers. Based on this design, the features from different stages and sensors are integrated for a MLP-based high-level learning. In the training phase, transfer learning is employed to infer the features learned from generic dataset to remote sensing data. The proposed FSN is evaluated by using the International Society for Photogrammetry and Remote Sensing (ISPRS) Potsdam and Vaihingen 2D Semantic Labeling datasets. Experimental results demonstrate that the proposed framework can bring considerable improvement to other related networks. [ABSTRACT FROM AUTHOR]